Readers' eye movements were recorded as they read sentences containing lexically ambiguous words or unambiguous control words. The ambiguous words had either two equally likely interpretations (equibiased words) or one highly dominant interpretation (non-equibiased words). On half the trials a preceding clause disambiguated the ambiguous word, while in the remaining trials the disambiguating clause followed the target word. Gaze durations on the target word and reading times on later parts of the sentences varied as a function of the type of target word (ambiguous or control), the type of ambiguous word (equibiased or non-equibiased), and the location of the disambiguating clause. The data from the study support a model in which more that one meaning of an ambiguous word is automatically activated, but degree of activation is influenced by frequency and prior context.